Big Data and CVD Benefits of Anti-Diabetes Medications
Dennis P. Scanlon, PhD: Bob, what do the CVD-REAL study results tell us about reducing cardiovascular events outside of the high-risk population with diabetes?
Robert Gabbay, MD, PhD, FACP: It’s the beginning of what I think we’ll see a lot more of, and that’s using big data to try to answer some of these questions. As we’ve heard, we don’t have a lot of answers. There’s a lot of hypotheses. Is it class effect? What’s the effect on congestive heart failure? We have some data, but not all of the questions are answered. There are a number of studies that are ongoing. Some, we’ll get results soon, but some will still take several years. What do we do in the meantime?
So, this was a retrospective analysis looking at a large data set to be able to identify one. Is there a signal for reduction of congestive heart failure across the class? It seemed as if there was. Again, the challenge with a retrospective analysis is you’re not randomizing people to therapy. There’s a whole host of potential confounders, but for a lot of the questions that we’ve brought up here today, we may need to depend on that kind of analysis to at least decide what to do with our patients that we see next week. So, I think there’s value in them, but we can’t put too much weight on them.
Dennis P. Scanlon, PhD: Are the studies at least attempting to control for some of those confounders, even if not in a randomized fashion?
Robert Gabbay, MD, PhD, FACP: Yes. I think the well-designed studies really do try to balance things as much as possible, again within the realm of the things that we know. Prior to the EMPA-REG OUTCOME trial, no one would have thought of matching, based on SGLT2 (sodium-glucose co-transporter 2), treatment. There will always be those surprises, but, yes, the goal is certainly the larger the data set, the more balancing one can do.
Silvio Inzucchi, MD: Although we should always take these observational studies with a grain of salt because of what you alluded to, which is really a channeling bias (certain patients get channeled to certain therapies-so we need to be cautious), the important aspect to CVD-REAL, is that it’s important because it included, as far as I recall, patients not necessarily on empagliflozin.
I think empagliflozin was actually represented with 1% or 2% of the participants, or the patient population. It was mostly driven by canagliflozin and dapagliflozin, and is a multi-national, observational study. They used propensity matching. I’m not a statistician but my statistical friends tell me that’s kind of the state-of-the-art way of doing these observational trials. You match patients for individual characteristics. They say it’s almost as good as a clinical trial. But the effects were as prominent as in EMPA-REG OUTCOME. There was a reduction in heart failure hospitalization and I believe they reported all-cause mortality
Zachary Bloomgarden, MD: They did. I was going to say that not only heart failure, but also mortality was decreased. The dilemma is, are enough variables measured so that the channeling bias can be overcome by the propensity matching algorithm? Are there certain things about that person who pushes his doctor to be on the new medicine which makes him or her have a better outcome? But yet, there are a number of questions that simply could never be addressed in a randomized controlled trial. I’m involved, now, in a study of sudden death in diabetes, which quantitatively is actually a pretty big deal. But because event rates are so low, you could never enroll enough people in a randomized controlled trial. So, you go to these propensity matching algorithms, you get large data sets, and maybe we can address the question of whether sulfonylureas are truly bad for you or if different ones are different.
Silvio Inzucchi, MD: Those drugs always look bad in these nonrandomized observational studies, but when they’re submitted to randomization in a clinical trial, they look either neutral or slightly beneficial. Now, the question is whether the patients who enroll in these randomized clinical trials are somehow different than those in the real world. They get better care in terms of closer observation, and that may not occur in the real world. I think it’s great when the randomized clinical trials and the observational data sets point in the same direction, but, when they don’t, I think it’s really confusing.
A big caveat comes from the DPP-4 (dipeptidyl peptidase 4) inhibitor experience. There were several large, observational, nonrandomized trials, retrospective, that suggested that the DPP-4 inhibitors had a dramatic effect on cardiovascular events. Yet, we now know that at least with 3 of the drugs that have been reported (sitagliptin, saxagliptin, and alogliptin), they are wonderfully neutral. In other words, no risk but clearly no benefit. So here you have a disconnect between those observational data. I’m not recalling whether they have propensity matching in those, but I think at least 1 did. Just a caution. We should wait for the randomized controlled trials.
Zachary Bloomgarden, MD: I put my foot in my mouth in print and said how wonderful they were. But there is a trial, CAROLINA, which will compare linagliptin with (I believe) glimepiride, head-to-head, in a high-risk cardiovascular disease population. I think in 2018 we’ll have an answer as to whether sulfonylureas, head-to-head with DPP-4s are bad, or if DPP-4 is good. So, let’s hang on to our hats for another installment.